摘要
为降低误差积累对线谱频率量化的影响,防止因此而导致语音质量下降,本文提出了基于分裂预测方法的新型混合式矢量量化结构。实验结果表明,该方法不仅比G.729的语音编解码算法明显降低了累积误差,平均谱失真在丢帧严重的情况下降低了达36%,而且提高了量化的透明度。
In this paper, a new scheme of vector quantization based on the split-predict method is presented, to reduce the effect of error propagation on Line Spectrum Frequency(LSF)quantization and avoid the voice quality decline. Simulation tests indicate that the new method is better than the standard technology used in G.729, making the average spectral distortion decline by 36% when badly losing frames, furthermore, improving the quantization transparency.
出处
《微计算机信息》
北大核心
2008年第4期263-264,299,共3页
Control & Automation
基金
国家"863"计划引导项目的支持(2004AA001390)
关键词
语音编码
线谱频率
分裂预测式矢量量化
Speech coding
Line spectrum frequency
Split-predict vector quantization
